The HYPERBOLA project aims to advance Ground Penetrating Radar (GPR) technology for robust, efficient, and precise subsurface reconstruction in complex, multilayered environments. The project focuses on estimating soil parameters and reconstructing layered scenes through a novel synergy of traditional algorithms and artificial intelligence (AI). It addresses key challenges in both target and non-target media reconstruction.

In target media, existing imaging algorithms such as back-projection (BP) and Stolt migration face limitations: a) Stolt migration inherently depends on depth, reducing flexibility; b) BP is computationally intensive, especially in layered media, due to the repetitive "delay-summation" process and the need for refractive point calculations; c) Both methods rely heavily on subjective focus metrics. To overcome these obstacles, HYPERBOLA introduces a vertex extraction-guided local BP algorithm and enhanced refraction point calculation techniques. These innovations significantly improve imaging efficiency and accuracy, particularly in complex, multilayered scenes.

In non-targhet media, full-wave inversion (FWI) and inverse scattering techniques are hampered by issues like low imaging efficiency, nonlinearity, and ill-conditioning. To address these challenges, the project combines the strengths of FWI and artificial neural networks (ANNs) to achieve high-fidelity reconstruction of layered media. By leveraging ANNs trained on the full-wave radar equation (Lambot et al., 2004), we aim to overcome traditional FWI limitations and enhance parameter estimation in environments without discrete objects.

Our synergistic approach integrates BP and FWI to leverage the strengths of both methods: BP provides efficient spatial imaging of target layers observed in B-scans, while FWI incorporates detailed wave propagation along layer interfaces from A-scans. This combination reduces the risk of local minima, decreases iteration counts, and improves parameter accuracy, making it ideal for advanced geophysical surveys and environmental research. Current efforts also explore AI-driven full-wave inversion for multilayered media, pushing the boundaries of GPR-based environmental monitoring.